Index: llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp =================================================================== --- llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp +++ llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp @@ -223,13 +223,14 @@ } SmallVector GCDRegs; - LLT GCDTy; - for (Register PartReg : PartRegs) - GCDTy = extractGCDType(GCDRegs, ResultTy, LeftoverTy, PartReg); - - for (Register PartReg : LeftoverRegs) - extractGCDType(GCDRegs, ResultTy, LeftoverTy, PartReg); + // Populate the first entry in GCDRegs separately in order to avoid + // recalculating GCDTy. + LLT GCDTy = extractGCDType(GCDRegs, ResultTy, LeftoverTy, PartRegs[0]); + for (auto PartReg : make_range(std::next(PartRegs.begin()), PartRegs.end())) + extractGCDType(GCDRegs, GCDTy, PartReg); + for (auto PartReg : LeftoverRegs) + extractGCDType(GCDRegs, GCDTy, PartReg); LLT ResultLCMTy = buildLCMMergePieces(ResultTy, LeftoverTy, GCDTy, GCDRegs); buildWidenedRemergeToDst(DstReg, ResultLCMTy, GCDRegs); } Index: llvm/test/CodeGen/AArch64/GlobalISel/legalize-constant.mir =================================================================== --- llvm/test/CodeGen/AArch64/GlobalISel/legalize-constant.mir +++ llvm/test/CodeGen/AArch64/GlobalISel/legalize-constant.mir @@ -1,31 +1,11 @@ # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py -# RUN: llc -O0 -run-pass=legalizer %s -o - | FileCheck %s +# RUN: llc -O0 -mtriple aarch64 -run-pass=legalizer -verify-machineinstrs %s -o - | FileCheck %s ---- | - target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128" - target triple = "aarch64--" - define void @test_constant() { - entry: - ret void - } - define void @test_fconstant() { - entry: - ret void - } -... - ---- name: test_constant registers: - - { id: 0, class: _ } - - { id: 1, class: _ } - - { id: 2, class: _ } - - { id: 3, class: _ } - - { id: 4, class: _ } - - { id: 5, class: _ } body: | - bb.0.entry: - + bb.0: + liveins: $x0 ; CHECK-LABEL: name: test_constant ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 ; CHECK: $w0 = COPY [[C]](s32) @@ -39,32 +19,28 @@ ; CHECK: $x0 = COPY [[C3]](s64) ; CHECK: [[C4:%[0-9]+]]:_(s64) = G_CONSTANT i64 0 ; CHECK: $x0 = COPY [[C4]](s64) - %0(s1) = G_CONSTANT i1 0 + %0:_(s1) = G_CONSTANT i1 0 %6:_(s32) = G_ANYEXT %0 $w0 = COPY %6 - %1(s8) = G_CONSTANT i8 42 + %1:_(s8) = G_CONSTANT i8 42 %7:_(s32) = G_ANYEXT %1 $w0 = COPY %7 - %2(s16) = G_CONSTANT i16 65535 + %2:_(s16) = G_CONSTANT i16 65535 %8:_(s32) = G_ANYEXT %2 $w0 = COPY %8 - %3(s32) = G_CONSTANT i32 -1 + %3:_(s32) = G_CONSTANT i32 -1 $w0 = COPY %3 - %4(s64) = G_CONSTANT i64 1 + %4:_(s64) = G_CONSTANT i64 1 $x0 = COPY %4 - %5(s64) = G_CONSTANT i64 0 + %5:_(s64) = G_CONSTANT i64 0 $x0 = COPY %5 ... --- name: test_fconstant -registers: - - { id: 0, class: _ } - - { id: 1, class: _ } - - { id: 2, class: _ } body: | - bb.0.entry: - + bb.0: + liveins: $x0 ; CHECK-LABEL: name: test_fconstant ; CHECK: [[C:%[0-9]+]]:_(s32) = G_FCONSTANT float 1.000000e+00 ; CHECK: $w0 = COPY [[C]](s32) @@ -74,11 +50,76 @@ ; CHECK: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[C2]](s32) ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16) ; CHECK: $w0 = COPY [[ANYEXT]](s32) - %0(s32) = G_FCONSTANT float 1.0 + %0:_(s32) = G_FCONSTANT float 1.0 $w0 = COPY %0 - %1(s64) = G_FCONSTANT double 2.0 + %1:_(s64) = G_FCONSTANT double 2.0 $x0 = COPY %1 - %2(s16) = G_FCONSTANT half 0.0 + %2:_(s16) = G_FCONSTANT half 0.0 %3:_(s32) = G_ANYEXT %2 $w0 = COPY %3 ... + +--- +name: s792 +registers: +body: | + bb.0: + liveins: $x0 + ; CHECK-LABEL: name: s792 + ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0 + ; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0 + ; CHECK: [[UV:%[0-9]+]]:_(s8), [[UV1:%[0-9]+]]:_(s8), [[UV2:%[0-9]+]]:_(s8), [[UV3:%[0-9]+]]:_(s8), [[UV4:%[0-9]+]]:_(s8), [[UV5:%[0-9]+]]:_(s8), [[UV6:%[0-9]+]]:_(s8), [[UV7:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[C]](s64) + ; CHECK: [[UV8:%[0-9]+]]:_(s8), [[UV9:%[0-9]+]]:_(s8), [[UV10:%[0-9]+]]:_(s8), [[UV11:%[0-9]+]]:_(s8), [[UV12:%[0-9]+]]:_(s8), [[UV13:%[0-9]+]]:_(s8), [[UV14:%[0-9]+]]:_(s8), [[UV15:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[C]](s64) + ; CHECK: [[UV16:%[0-9]+]]:_(s8), [[UV17:%[0-9]+]]:_(s8), [[UV18:%[0-9]+]]:_(s8), [[UV19:%[0-9]+]]:_(s8), [[UV20:%[0-9]+]]:_(s8), [[UV21:%[0-9]+]]:_(s8), [[UV22:%[0-9]+]]:_(s8), [[UV23:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[C]](s64) + ; CHECK: [[UV24:%[0-9]+]]:_(s8), [[UV25:%[0-9]+]]:_(s8), [[UV26:%[0-9]+]]:_(s8), [[UV27:%[0-9]+]]:_(s8), [[UV28:%[0-9]+]]:_(s8), [[UV29:%[0-9]+]]:_(s8), [[UV30:%[0-9]+]]:_(s8), [[UV31:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[C]](s64) + ; CHECK: [[UV32:%[0-9]+]]:_(s8), [[UV33:%[0-9]+]]:_(s8), [[UV34:%[0-9]+]]:_(s8), [[UV35:%[0-9]+]]:_(s8), [[UV36:%[0-9]+]]:_(s8), [[UV37:%[0-9]+]]:_(s8), [[UV38:%[0-9]+]]:_(s8), [[UV39:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[C]](s64) + ; CHECK: [[UV40:%[0-9]+]]:_(s8), [[UV41:%[0-9]+]]:_(s8), [[UV42:%[0-9]+]]:_(s8), [[UV43:%[0-9]+]]:_(s8), [[UV44:%[0-9]+]]:_(s8), [[UV45:%[0-9]+]]:_(s8), [[UV46:%[0-9]+]]:_(s8), [[UV47:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[C]](s64) + ; CHECK: [[UV48:%[0-9]+]]:_(s8), [[UV49:%[0-9]+]]:_(s8), [[UV50:%[0-9]+]]:_(s8), [[UV51:%[0-9]+]]:_(s8), [[UV52:%[0-9]+]]:_(s8), [[UV53:%[0-9]+]]:_(s8), [[UV54:%[0-9]+]]:_(s8), [[UV55:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[C]](s64) + ; CHECK: [[UV56:%[0-9]+]]:_(s8), [[UV57:%[0-9]+]]:_(s8), [[UV58:%[0-9]+]]:_(s8), [[UV59:%[0-9]+]]:_(s8), [[UV60:%[0-9]+]]:_(s8), [[UV61:%[0-9]+]]:_(s8), [[UV62:%[0-9]+]]:_(s8), [[UV63:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[C]](s64) + ; CHECK: [[UV64:%[0-9]+]]:_(s8), [[UV65:%[0-9]+]]:_(s8), [[UV66:%[0-9]+]]:_(s8), [[UV67:%[0-9]+]]:_(s8), [[UV68:%[0-9]+]]:_(s8), [[UV69:%[0-9]+]]:_(s8), [[UV70:%[0-9]+]]:_(s8), [[UV71:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[C]](s64) + ; CHECK: [[UV72:%[0-9]+]]:_(s8), [[UV73:%[0-9]+]]:_(s8), [[UV74:%[0-9]+]]:_(s8), [[UV75:%[0-9]+]]:_(s8), [[UV76:%[0-9]+]]:_(s8), [[UV77:%[0-9]+]]:_(s8), [[UV78:%[0-9]+]]:_(s8), [[UV79:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[C]](s64) + ; CHECK: [[UV80:%[0-9]+]]:_(s8), [[UV81:%[0-9]+]]:_(s8), [[UV82:%[0-9]+]]:_(s8), [[UV83:%[0-9]+]]:_(s8), [[UV84:%[0-9]+]]:_(s8), [[UV85:%[0-9]+]]:_(s8), [[UV86:%[0-9]+]]:_(s8), [[UV87:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[C]](s64) + ; CHECK: [[UV88:%[0-9]+]]:_(s8), [[UV89:%[0-9]+]]:_(s8), [[UV90:%[0-9]+]]:_(s8), [[UV91:%[0-9]+]]:_(s8), [[UV92:%[0-9]+]]:_(s8), [[UV93:%[0-9]+]]:_(s8), [[UV94:%[0-9]+]]:_(s8), [[UV95:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[C]](s64) + ; CHECK: [[UV96:%[0-9]+]]:_(s8), [[UV97:%[0-9]+]]:_(s8), [[UV98:%[0-9]+]]:_(s8), [[UV99:%[0-9]+]]:_(s8) = G_UNMERGE_VALUES [[C1]](s32) + ; CHECK: [[MV:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV]](s8), [[UV1]](s8), [[UV2]](s8) + ; CHECK: [[MV1:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV3]](s8), [[UV4]](s8), [[UV5]](s8) + ; CHECK: [[MV2:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV6]](s8), [[UV7]](s8), [[UV8]](s8) + ; CHECK: [[MV3:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV9]](s8), [[UV10]](s8), [[UV11]](s8) + ; CHECK: [[MV4:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV12]](s8), [[UV13]](s8), [[UV14]](s8) + ; CHECK: [[MV5:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV15]](s8), [[UV16]](s8), [[UV17]](s8) + ; CHECK: [[MV6:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV18]](s8), [[UV19]](s8), [[UV20]](s8) + ; CHECK: [[MV7:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV21]](s8), [[UV22]](s8), [[UV23]](s8) + ; CHECK: [[MV8:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV24]](s8), [[UV25]](s8), [[UV26]](s8) + ; CHECK: [[MV9:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV27]](s8), [[UV28]](s8), [[UV29]](s8) + ; CHECK: [[MV10:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV30]](s8), [[UV31]](s8), [[UV32]](s8) + ; CHECK: [[MV11:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV33]](s8), [[UV34]](s8), [[UV35]](s8) + ; CHECK: [[MV12:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV36]](s8), [[UV37]](s8), [[UV38]](s8) + ; CHECK: [[MV13:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV39]](s8), [[UV40]](s8), [[UV41]](s8) + ; CHECK: [[MV14:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV42]](s8), [[UV43]](s8), [[UV44]](s8) + ; CHECK: [[MV15:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV45]](s8), [[UV46]](s8), [[UV47]](s8) + ; CHECK: [[MV16:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV48]](s8), [[UV49]](s8), [[UV50]](s8) + ; CHECK: [[MV17:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV51]](s8), [[UV52]](s8), [[UV53]](s8) + ; CHECK: [[MV18:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV54]](s8), [[UV55]](s8), [[UV56]](s8) + ; CHECK: [[MV19:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV57]](s8), [[UV58]](s8), [[UV59]](s8) + ; CHECK: [[MV20:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV60]](s8), [[UV61]](s8), [[UV62]](s8) + ; CHECK: [[MV21:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV63]](s8), [[UV64]](s8), [[UV65]](s8) + ; CHECK: [[MV22:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV66]](s8), [[UV67]](s8), [[UV68]](s8) + ; CHECK: [[MV23:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV69]](s8), [[UV70]](s8), [[UV71]](s8) + ; CHECK: [[MV24:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV72]](s8), [[UV73]](s8), [[UV74]](s8) + ; CHECK: [[MV25:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV75]](s8), [[UV76]](s8), [[UV77]](s8) + ; CHECK: [[MV26:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV78]](s8), [[UV79]](s8), [[UV80]](s8) + ; CHECK: [[MV27:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV81]](s8), [[UV82]](s8), [[UV83]](s8) + ; CHECK: [[MV28:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV84]](s8), [[UV85]](s8), [[UV86]](s8) + ; CHECK: [[MV29:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV87]](s8), [[UV88]](s8), [[UV89]](s8) + ; CHECK: [[MV30:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV90]](s8), [[UV91]](s8), [[UV92]](s8) + ; CHECK: [[MV31:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV93]](s8), [[UV94]](s8), [[UV95]](s8) + ; CHECK: [[MV32:%[0-9]+]]:_(s24) = G_MERGE_VALUES [[UV96]](s8), [[UV97]](s8), [[UV98]](s8) + ; CHECK: %cst:_(s792) = G_MERGE_VALUES [[MV]](s24), [[MV1]](s24), [[MV2]](s24), [[MV3]](s24), [[MV4]](s24), [[MV5]](s24), [[MV6]](s24), [[MV7]](s24), [[MV8]](s24), [[MV9]](s24), [[MV10]](s24), [[MV11]](s24), [[MV12]](s24), [[MV13]](s24), [[MV14]](s24), [[MV15]](s24), [[MV16]](s24), [[MV17]](s24), [[MV18]](s24), [[MV19]](s24), [[MV20]](s24), [[MV21]](s24), [[MV22]](s24), [[MV23]](s24), [[MV24]](s24), [[MV25]](s24), [[MV26]](s24), [[MV27]](s24), [[MV28]](s24), [[MV29]](s24), [[MV30]](s24), [[MV31]](s24), [[MV32]](s24) + ; CHECK: %trunc:_(s32) = G_TRUNC %cst(s792) + ; CHECK: $w0 = COPY %trunc(s32) + ; CHECK: RET_ReallyLR implicit $w0 + %cst:_(s792) = G_CONSTANT i792 0 + %trunc:_(s32) = G_TRUNC %cst(s792) + $w0 = COPY %trunc(s32) + RET_ReallyLR implicit $w0 +...